Computational Methods for Detecting Multi-gene Relationships in Gene Expression Data

基因共表达在过去几年中作为一种强大的生物信息学工具出现。共表达基因可能参与相同的生物学过程，因此常使用共表达网络来研究基因功能。目前，共表达基因对的检测、分析和聚类方法已经得到了很好的发展。然而，两两共表达显然过于简单化，无法描述基因之间的复杂关系，因为这些关系可能涉及多个基因，并且可能因生物环境而异。为了解决这一限制，我们建议开发用于检测和分析基因表达数据中的多基因关系的计算方法，并使用这些方法来解决分子生物学中的问题。这些问题将包括基因调控网络的重建、基因调控调节剂的检测以及基因功能预测工具的改进。

Gene coexpression has emerged over the last several years as a powerful bioinformatics tool. Coexpressed genes may be involved in the same biological process, and thus coexpression networks are often used to investigate gene function. Methods for detecting, analyzing and clustering pairs of coexpressed genes are now well developed. Pairwise coexpression, however, is clearly too simplistic to describe the complex relationships between genes, since these relationships may involve multiple genes and can vary depending on the biological context. To address this limitation, we propose to develop computational methods for detecting and analyzing multi-gene relationships in gene expression data, and to use these methods to solve problems in molecular biology. These problems will include the reconstruction of genetic regulatory networks, the detection of modulators of gene regulation, and improved tools for gene function prediction. One of the main difficulties in detecting multi-gene relationships is the vast number of possible relationships which need to be considered. This is especially true in organisms with a large number of genes, such as plants and mammals. In such cases, separating the true relationships from the possible relationships poses challenging computational and statistical problems. To address these problems, we shall first develop methods of low computational complexity for detecting relatively simple multi-gene relationships. To avoid overfitting the data, we will develop methods for accurately estimating false positive rates. We will also develop heuristics for speeding up the computations while missing as few significant relationships as possible. To validate the methods, we will test them on both real and simulated gene expression data. Building upon these results, we will develop methods for detecting more-complex relationships involving larger numbers of genes.